[x64] add Absps/d and Negps/d macro

[x64] add Absps/d and Negps/d macro

BUG=

Committed: https://crrev.com/b973a77775ae258820578838393d738c6d235f65
Cr-Commit-Position: refs/heads/master@{#37937}

[Weiliang, 2016-07-18 04:43:54]

weiliang.lin@intel.com changed reviewers:
+ bbudge@chromium.org, bmeurer@chromium.org, gdeepti@google.com

[Weiliang, 2016-07-18 04:43:55]

PTAL

[Weiliang, 2016-07-18 04:45:50]

Description was changed from

==========
x64: add Absps and Negps macro

BUG=
==========

to

==========
[x64] add Absps and Negps macro

BUG=
==========

[Benedikt Meurer, 2016-07-18 04:47:52]

https://codereview.chromium.org/2161513002/diff/1/src/x64/macro-assembler-x64.cc
File src/x64/macro-assembler-x64.cc (right):

https://codereview.chromium.org/2161513002/diff/1/src/x64/macro-assembler-x64...
src/x64/macro-assembler-x64.cc:2986: Set(kScratchRegister,
reinterpret_cast<intptr_t>(&float_absolute_constant));
Can we somehow put this somewhere near the roots array so we can use the
ExternalOperand mechanism to avoid this load?

Also can you add Abspd and Negpd versions as well?

[bbudge, 2016-07-18 14:08:09]

https://codereview.chromium.org/2161513002/diff/1/src/x64/macro-assembler-x64.cc
File src/x64/macro-assembler-x64.cc (right):

https://codereview.chromium.org/2161513002/diff/1/src/x64/macro-assembler-x64...
src/x64/macro-assembler-x64.cc:2999: Xorps(dst, Operand(kScratchRegister, 0));
It seems like we could avoid the memory access with AVX:

vpxor kScratch, kScratch, kScratch
vsubps dst, kScratch, dst

Is there a reason we can't use 0 - dst?

A similar trick can be done with three instructions for SSE.

[Weiliang, 2016-07-19 08:24:01]

On 2016/07/18 04:47:52, Benedikt Meurer wrote:
>
https://codereview.chromium.org/2161513002/diff/1/src/x64/macro-assembler-x64.cc
> File src/x64/macro-assembler-x64.cc (right):
> 
>
https://codereview.chromium.org/2161513002/diff/1/src/x64/macro-assembler-x64...
> src/x64/macro-assembler-x64.cc:2986: Set(kScratchRegister,
> reinterpret_cast<intptr_t>(&float_absolute_constant));
> Can we somehow put this somewhere near the roots array so we can use the
> ExternalOperand mechanism to avoid this load?
> 
> Also can you add Abspd and Negpd versions as well?

Abspd and Negpd are provided. 
And the constants are moved to assembler.cc just like some double constant. Is
it OK?

[Weiliang, 2016-07-19 08:28:01]

On 2016/07/18 14:08:09, bbudge wrote:
>
https://codereview.chromium.org/2161513002/diff/1/src/x64/macro-assembler-x64.cc
> File src/x64/macro-assembler-x64.cc (right):
> 
>
https://codereview.chromium.org/2161513002/diff/1/src/x64/macro-assembler-x64...
> src/x64/macro-assembler-x64.cc:2999: Xorps(dst, Operand(kScratchRegister, 0));
> It seems like we could avoid the memory access with AVX:
> 
> vpxor kScratch, kScratch, kScratch
> vsubps dst, kScratch, dst
> 
> Is there a reason we can't use 0 - dst?
> 
> A similar trick can be done with three instructions for SSE.

We used this implementation in our previous prototype. I am not sure which one
is always better than the other, but how about making the choice lazily at the
performance tuning stage.

[bbudge, 2016-07-19 11:23:38]

On 2016/07/19 08:28:01, Weiliang wrote:
> On 2016/07/18 14:08:09, bbudge wrote:
> >
>
https://codereview.chromium.org/2161513002/diff/1/src/x64/macro-assembler-x64.cc
> > File src/x64/macro-assembler-x64.cc (right):
> > 
> >
>
https://codereview.chromium.org/2161513002/diff/1/src/x64/macro-assembler-x64...
> > src/x64/macro-assembler-x64.cc:2999: Xorps(dst, Operand(kScratchRegister,
0));
> > It seems like we could avoid the memory access with AVX:
> > 
> > vpxor kScratch, kScratch, kScratch
> > vsubps dst, kScratch, dst
> > 
> > Is there a reason we can't use 0 - dst?
> > 
> > A similar trick can be done with three instructions for SSE.
> 
> We used this implementation in our previous prototype. I am not sure which one
> is always better than the other, but how about making the choice lazily at the
> performance tuning stage.

OK, just wanted to mention it.

[Weiliang, 2016-07-21 09:01:28]

Description was changed from

==========
[x64] add Absps and Negps macro

BUG=
==========

to

==========
[x64] add Absps/d and Negps/d macro

BUG=
==========

[Weiliang, 2016-07-21 09:01:39]

PTAL

[Benedikt Meurer, 2016-07-21 09:03:06]

LGTM.

[Weiliang, 2016-07-21 09:40:07]

The CQ bit was checked by weiliang.lin@intel.com

[commit-bot: I haz the power, 2016-07-21 09:40:18]

CQ is trying da patch. Follow status at
 
https://chromium-cq-status.appspot.com/v2/patch-status/codereview.chromium.or...

[commit-bot: I haz the power, 2016-07-21 11:01:50]

Description was changed from

==========
[x64] add Absps/d and Negps/d macro

BUG=
==========

to

==========
[x64] add Absps/d and Negps/d macro

BUG=
==========

[commit-bot: I haz the power, 2016-07-21 11:01:51]

Committed patchset #2 (id:20001)

[commit-bot: I haz the power, 2016-07-21 11:02:51]

Description was changed from

==========
[x64] add Absps/d and Negps/d macro

BUG=
==========

to

==========
[x64] add Absps/d and Negps/d macro

BUG=

Committed: https://crrev.com/b973a77775ae258820578838393d738c6d235f65
Cr-Commit-Position: refs/heads/master@{#37937}
==========

[commit-bot: I haz the power, 2016-07-21 11:02:53]

Patchset 2 (id:??) landed as
https://crrev.com/b973a77775ae258820578838393d738c6d235f65
Cr-Commit-Position: refs/heads/master@{#37937}